Applying synergy metrics to combination screening data: agreements, disagreements and pitfalls

Area of Science:

• Oncology
• Pharmacology
• Computational Biology

Background:

• Synergistic drug combinations are essential for overcoming cancer monotherapy resistance.
• High-throughput screening is used to identify effective drug combinations.
• Synergy quantification relies on various models with differing assumptions.

Purpose of the Study:

• To compare the behavior of four common synergy models: Loewe additivity, Bliss independence, highest single agent, and zero interaction potency.
• To evaluate model agreement and disagreement using a real-world cancer combination screen dataset.
• To investigate how model assumptions and experimental factors influence synergy quantification.

Main Methods:

• Utilized the Merck oncology combination screen dataset.
• Applied four distinct synergy quantification models: Loewe additivity, Bliss independence, highest single agent, and zero interaction potency.
• Performed statistical analysis to assess model concordance (Pearson's r, Spearman's ρ) and identify discrepancies.

Main Results:

• Observed moderate concordance between models (Pearson's r >0.32, Spearman's ρ>0.34).
• Identified multiple instances of strong disagreement between the four synergy models.
• Disagreements were attributed to variations in tested concentrations, maximum response values, and median effective concentrations.

Conclusions:

• Different synergy models exhibit significant disagreements, impacting the identification of effective cancer drug combinations.
• Experimental factors such as drug concentration and response metrics critically influence synergy model outcomes.
• Careful consideration of model assumptions and experimental design is necessary for accurate synergy assessment in cancer drug discovery.